Feature engineering on prediction data I want to predict with

I want tp call m.predict() on some data, but that data has to fit the dimensions of X_train, but after a lot of feature engineering it is difficult to shape the new input to fit the model.

If the raw data looked like this (dataframe.info()) :

> Data columns (total 5 columns):
> date            58764 non-null object
> vehicleCount    58764 non-null int64
> id              58764 non-null int64
> totalSpaces     58764 non-null int64
> garageCode      58764 non-null object
> dtypes: int64(3), object(2)

It might look like this at training time:

> Data columns (total 26 columns):
> Hourofday                     47764 non-null int64
> Minutesofhour                 47764 non-null int64
> Year                          47764 non-null int64
> Month                         47764 non-null int64
> Week                          47764 non-null int64
> Day                           47764 non-null int64
> Dayofweek                     47764 non-null int64
> Dayofyear                     47764 non-null int64
> Is_month_end                  47764 non-null bool
> Is_month_start                47764 non-null bool
> Is_quarter_end                47764 non-null bool
> Is_quarter_start              47764 non-null bool
> Is_year_end                   47764 non-null bool
> Is_year_start                 47764 non-null bool
 [...]
 (11 one-hot columns, thats another issue as well)
 [...]
> dtypes: bool(6), int64(8), uint8(12)

Are there any good practices to shape the new input?

you need to train a model for your new features.